RDMA/cxgb3: Doorbell overflow avoidance and recovery

T3 hardware doorbell FIFO overflows can cause application stalls due
to lost doorbell ring events.  This has been seen when running large
NP IMB alltoall MPI jobs.  The T3 hardware supports an xon/xoff-type
flow control mechanism to help avoid overflowing the HW doorbell FIFO.

This patch uses these interrupts to disable RDMA QP doorbell rings
when we near an overflow condition, and then turn them back on (and
ring all the active QP doorbells) when when the doorbell FIFO empties
out.  In addition if an doorbell ring is dropped by the hardware, the
code will now recover.

Design:

cxgb3:
- enable these DB interrupts
- in the interrupt handler, schedule work tasks to call the ULPs event
  handlers with the new events.
- ring all the qset txqs when an overflow is detected.

iw_cxgb3:
- disable db ringing on all active qps when we get the DB_FULL event
- enable db ringing on all active qps and ring all active dbs when we get
  the DB_EMPTY event
- On DB_DROP event:
       - disable db rings in the event handler
       - delay-schedule a work task which rings and enables the dbs on
         all active qps.
- in post_send and post_recv logic, don't ring the db if it's disabled.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

1 files changed, 56 insertions, 1 deletions

diff --git a/drivers/net/cxgb3/cxgb3_main.c b/drivers/net/cxgb3/cxgb3_main.c
index 89bec9c3c14..37945fce7fa 100644
--- a/drivers/net/cxgb3/cxgb3_main.c
+++ b/drivers/net/cxgb3/cxgb3_main.c
@@ -45,6 +45,7 @@
 #include <linux/firmware.h>
 #include <linux/log2.h>
 #include <linux/stringify.h>
+#include <linux/sched.h>
 #include <asm/uaccess.h>
 
 #include "common.h"
@@ -140,7 +141,7 @@ MODULE_PARM_DESC(ofld_disable, "whether to enable offload at init time or not");
  * will block keventd as it needs the rtnl lock, and we'll deadlock waiting
  * for our work to complete.  Get our own work queue to solve this.
  */
-static struct workqueue_struct *cxgb3_wq;
+struct workqueue_struct *cxgb3_wq;
 
 /**
  *	link_report - show link status and link speed/duplex
@@ -590,6 +591,19 @@ static void setup_rss(struct adapter *adap)
 		      V_RRCPLCPUSIZE(6) | F_HASHTOEPLITZ, cpus, rspq_map);
 }
 
+static void ring_dbs(struct adapter *adap)
+{
+	int i, j;
+
+	for (i = 0; i < SGE_QSETS; i++) {
+		struct sge_qset *qs = &adap->sge.qs[i];
+
+		if (qs->adap)
+			for (j = 0; j < SGE_TXQ_PER_SET; j++)
+				t3_write_reg(adap, A_SG_KDOORBELL, F_SELEGRCNTX | V_EGRCNTX(qs->txq[j].cntxt_id));
+	}
+}
+
 static void init_napi(struct adapter *adap)
 {
 	int i;
@@ -2754,6 +2768,42 @@ static void t3_adap_check_task(struct work_struct *work)
 	spin_unlock_irq(&adapter->work_lock);
 }
 
+static void db_full_task(struct work_struct *work)
+{
+	struct adapter *adapter = container_of(work, struct adapter,
+					       db_full_task);
+
+	cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_FULL, 0);
+}
+
+static void db_empty_task(struct work_struct *work)
+{
+	struct adapter *adapter = container_of(work, struct adapter,
+					       db_empty_task);
+
+	cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_EMPTY, 0);
+}
+
+static void db_drop_task(struct work_struct *work)
+{
+	struct adapter *adapter = container_of(work, struct adapter,
+					       db_drop_task);
+	unsigned long delay = 1000;
+	unsigned short r;
+
+	cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_DROP, 0);
+
+	/*
+	 * Sleep a while before ringing the driver qset dbs.
+	 * The delay is between 1000-2023 usecs.
+	 */
+	get_random_bytes(&r, 2);
+	delay += r & 1023;
+	set_current_state(TASK_UNINTERRUPTIBLE);
+	schedule_timeout(usecs_to_jiffies(delay));
+	ring_dbs(adapter);
+}
+
 /*
  * Processes external (PHY) interrupts in process context.
  */
@@ -3222,6 +3272,11 @@ static int __devinit init_one(struct pci_dev *pdev,
 	INIT_LIST_HEAD(&adapter->adapter_list);
 	INIT_WORK(&adapter->ext_intr_handler_task, ext_intr_task);
 	INIT_WORK(&adapter->fatal_error_handler_task, fatal_error_task);
+
+	INIT_WORK(&adapter->db_full_task, db_full_task);
+	INIT_WORK(&adapter->db_empty_task, db_empty_task);
+	INIT_WORK(&adapter->db_drop_task, db_drop_task);
+
 	INIT_DELAYED_WORK(&adapter->adap_check_task, t3_adap_check_task);
 
 	for (i = 0; i < ai->nports0 + ai->nports1; ++i) {